aom_dsp/flow_estimation/x86/disflow_sse4.c - aom - Git at Google

 /*
  * Copyright (c) 2024, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
  * was not distributed with this source code in the LICENSE file, you can
  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
  * Media Patent License 1.0 was not distributed with this source code in the
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */

 #include <assert.h>
 #include <math.h>
 #include <smmintrin.h>

 #include "aom_dsp/aom_dsp_common.h"
 #include "aom_dsp/flow_estimation/disflow.h"
 #include "aom_dsp/x86/synonyms.h"

 #include "config/aom_dsp_rtcd.h"

 #if DISFLOW_PATCH_SIZE != 8
 #error "Need to change disflow_sse4.c if DISFLOW_PATCH_SIZE != 8"
 #endif

 // Compute horizontal and vertical kernels and return them packed into a
 // register. The coefficient ordering is:
 //   h0, h1, v0, v1, h2, h3, v2, v3
 // This is chosen because it takes less work than fully separating the kernels,
 // but it is separated enough that we can pick out each coefficient pair in the
 // main compute_flow_at_point function
 static INLINE __m128i compute_cubic_kernels(double u, double v) {
   const __m128d x = _mm_set_pd(v, u);

   const __m128d x2 = _mm_mul_pd(x, x);
   const __m128d x3 = _mm_mul_pd(x2, x);

   // Macro to multiply a value v by a constant coefficient c
 #define MULC(c, v) _mm_mul_pd(_mm_set1_pd(c), v)

   // Compute floating-point kernel
   // Note: To ensure results are bit-identical to the C code, we need to perform
   // exactly the same sequence of operations here as in the C code.
   __m128d k0 = _mm_sub_pd(_mm_add_pd(MULC(-0.5, x), x2), MULC(0.5, x3));
   __m128d k1 =
       _mm_add_pd(_mm_sub_pd(_mm_set1_pd(1.0), MULC(2.5, x2)), MULC(1.5, x3));
   __m128d k2 =
       _mm_sub_pd(_mm_add_pd(MULC(0.5, x), MULC(2.0, x2)), MULC(1.5, x3));
   __m128d k3 = _mm_add_pd(MULC(-0.5, x2), MULC(0.5, x3));
 #undef MULC

   // Integerize
   __m128d prec = _mm_set1_pd((double)(1 << DISFLOW_INTERP_BITS));

   k0 = _mm_round_pd(_mm_mul_pd(k0, prec),
                     _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
   k1 = _mm_round_pd(_mm_mul_pd(k1, prec),
                     _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
   k2 = _mm_round_pd(_mm_mul_pd(k2, prec),
                     _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
   k3 = _mm_round_pd(_mm_mul_pd(k3, prec),
                     _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);

   const __m128i c0 = _mm_cvtpd_epi32(k0);
   const __m128i c1 = _mm_cvtpd_epi32(k1);
   const __m128i c2 = _mm_cvtpd_epi32(k2);
   const __m128i c3 = _mm_cvtpd_epi32(k3);

   // Rearrange results and convert down to 16 bits, giving the target output
   // ordering
   const __m128i c01 = _mm_unpacklo_epi32(c0, c1);
   const __m128i c23 = _mm_unpacklo_epi32(c2, c3);
   return _mm_packs_epi32(c01, c23);
 }

 // Compare two regions of width x height pixels, one rooted at position
 // (x, y) in src and the other at (x + u, y + v) in ref.
 // This function returns the sum of squared pixel differences between
 // the two regions.
 //
 // TODO(rachelbarker): Test speed/quality impact of using bilinear interpolation
 // instad of bicubic interpolation
 static INLINE void compute_flow_vector(const uint8_t *src, const uint8_t *ref,
                                        int width, int height, int stride, int x,
                                        int y, double u, double v,
                                        const int16_t *dx, const int16_t *dy,
                                        int *b) {
   // This function is written to do 8x8 convolutions only
   assert(DISFLOW_PATCH_SIZE == 8);

   // Accumulate 4 32-bit partial sums for each element of b
   // These will be flattened at the end.
   __m128i b0_acc = _mm_setzero_si128();
   __m128i b1_acc = _mm_setzero_si128();

   // Split offset into integer and fractional parts, and compute cubic
   // interpolation kernels
   const int u_int = (int)floor(u);
   const int v_int = (int)floor(v);
   const double u_frac = u - floor(u);
   const double v_frac = v - floor(v);

   const __m128i kernels = compute_cubic_kernels(u_frac, v_frac);

   // Storage for intermediate values between the two convolution directions
   DECLARE_ALIGNED(16, int16_t,
                   tmp_[DISFLOW_PATCH_SIZE * (DISFLOW_PATCH_SIZE + 3)]);
   int16_t *tmp = tmp_ + DISFLOW_PATCH_SIZE;  // Offset by one row

   // Clamp coordinates so that all pixels we fetch will remain within the
   // allocated border region, but allow them to go far enough out that
   // the border pixels' values do not change.
   // Since we are calculating an 8x8 block, the bottom-right pixel
   // in the block has coordinates (x0 + 7, y0 + 7). Then, the cubic
   // interpolation has 4 taps, meaning that the output of pixel
   // (x_w, y_w) depends on the pixels in the range
   // ([x_w - 1, x_w + 2], [y_w - 1, y_w + 2]).
   //
   // Thus the most extreme coordinates which will be fetched are
   // (x0 - 1, y0 - 1) and (x0 + 9, y0 + 9).
   const int x0 = clamp(x + u_int, -9, width);
   const int y0 = clamp(y + v_int, -9, height);

   // Horizontal convolution

   // Prepare the kernel vectors
   // We split the kernel into two vectors with kernel indices:
   // 0, 1, 0, 1, 0, 1, 0, 1, and
   // 2, 3, 2, 3, 2, 3, 2, 3
   __m128i h_kernel_01 = _mm_set1_epi32(_mm_extract_epi32(kernels, 0));
   __m128i h_kernel_23 = _mm_set1_epi32(_mm_extract_epi32(kernels, 2));

   __m128i round_const_h = _mm_set1_epi32(1 << (DISFLOW_INTERP_BITS - 6 - 1));

   for (int i = -1; i < DISFLOW_PATCH_SIZE + 2; ++i) {
     const int y_w = y0 + i;
     const uint8_t *ref_row = &ref[y_w * stride + (x0 - 1)];
     int16_t *tmp_row = &tmp[i * DISFLOW_PATCH_SIZE];

     // Load this row of pixels.
     // For an 8x8 patch, we need to load the 8 image pixels + 3 extras,
     // for a total of 11 pixels. Here we load 16 pixels, but only use
     // the first 11.
     __m128i row = _mm_loadu_si128((__m128i *)ref_row);

     // Expand pixels to int16s
     __m128i px_0to7_i16 = _mm_cvtepu8_epi16(row);
     __m128i px_4to10_i16 = _mm_cvtepu8_epi16(_mm_srli_si128(row, 4));

     // Compute first four outputs
     // input pixels 0, 1, 1, 2, 2, 3, 3, 4
     // * kernel     0, 1, 0, 1, 0, 1, 0, 1
     __m128i px0 =
         _mm_unpacklo_epi16(px_0to7_i16, _mm_srli_si128(px_0to7_i16, 2));
     // input pixels 2, 3, 3, 4, 4, 5, 5, 6
     // * kernel     2, 3, 2, 3, 2, 3, 2, 3
     __m128i px1 = _mm_unpacklo_epi16(_mm_srli_si128(px_0to7_i16, 4),
                                      _mm_srli_si128(px_0to7_i16, 6));
     // Convolve with kernel and sum 2x2 boxes to form first 4 outputs
     __m128i sum0 = _mm_add_epi32(_mm_madd_epi16(px0, h_kernel_01),
                                  _mm_madd_epi16(px1, h_kernel_23));

     __m128i out0 = _mm_srai_epi32(_mm_add_epi32(sum0, round_const_h),
                                   DISFLOW_INTERP_BITS - 6);

     // Compute second four outputs
     __m128i px2 =
         _mm_unpacklo_epi16(px_4to10_i16, _mm_srli_si128(px_4to10_i16, 2));
     __m128i px3 = _mm_unpacklo_epi16(_mm_srli_si128(px_4to10_i16, 4),
                                      _mm_srli_si128(px_4to10_i16, 6));
     __m128i sum1 = _mm_add_epi32(_mm_madd_epi16(px2, h_kernel_01),
                                  _mm_madd_epi16(px3, h_kernel_23));

     // Round by just enough bits that the result is
     // guaranteed to fit into an i16. Then the next stage can use 16 x 16 -> 32
     // bit multiplies, which should be a fair bit faster than 32 x 32 -> 32
     // as it does now
     // This means shifting down so we have 6 extra bits, for a maximum value
     // of +18360, which can occur if u_frac == 0.5 and the input pixels are
     // {0, 255, 255, 0}.
     __m128i out1 = _mm_srai_epi32(_mm_add_epi32(sum1, round_const_h),
                                   DISFLOW_INTERP_BITS - 6);

     _mm_storeu_si128((__m128i *)tmp_row, _mm_packs_epi32(out0, out1));
   }

   // Vertical convolution
   const int round_bits = DISFLOW_INTERP_BITS + 6 - DISFLOW_DERIV_SCALE_LOG2;
   __m128i round_const_v = _mm_set1_epi32(1 << (round_bits - 1));

   __m128i v_kernel_01 = _mm_set1_epi32(_mm_extract_epi32(kernels, 1));
   __m128i v_kernel_23 = _mm_set1_epi32(_mm_extract_epi32(kernels, 3));

   for (int i = 0; i < DISFLOW_PATCH_SIZE; ++i) {
     int16_t *tmp_row = &tmp[i * DISFLOW_PATCH_SIZE];

     // Load 4 rows of 8 x 16-bit values
     __m128i px0 = _mm_loadu_si128((__m128i *)(tmp_row - DISFLOW_PATCH_SIZE));
     __m128i px1 = _mm_loadu_si128((__m128i *)tmp_row);
     __m128i px2 = _mm_loadu_si128((__m128i *)(tmp_row + DISFLOW_PATCH_SIZE));
     __m128i px3 =
         _mm_loadu_si128((__m128i *)(tmp_row + 2 * DISFLOW_PATCH_SIZE));

     // We want to calculate px0 * v_kernel[0] + px1 * v_kernel[1] + ... ,
     // but each multiply expands its output to 32 bits. So we need to be
     // a little clever about how we do this
     __m128i sum0 = _mm_add_epi32(
         _mm_madd_epi16(_mm_unpacklo_epi16(px0, px1), v_kernel_01),
         _mm_madd_epi16(_mm_unpacklo_epi16(px2, px3), v_kernel_23));
     __m128i sum1 = _mm_add_epi32(
         _mm_madd_epi16(_mm_unpackhi_epi16(px0, px1), v_kernel_01),
         _mm_madd_epi16(_mm_unpackhi_epi16(px2, px3), v_kernel_23));

     __m128i sum0_rounded =
         _mm_srai_epi32(_mm_add_epi32(sum0, round_const_v), round_bits);
     __m128i sum1_rounded =
         _mm_srai_epi32(_mm_add_epi32(sum1, round_const_v), round_bits);

     __m128i warped = _mm_packs_epi32(sum0_rounded, sum1_rounded);
     __m128i src_pixels_u8 =
         _mm_loadl_epi64((__m128i *)&src[(y + i) * stride + x]);
     __m128i src_pixels = _mm_slli_epi16(_mm_cvtepu8_epi16(src_pixels_u8), 3);

     // Calculate delta from the target patch
     __m128i dt = _mm_sub_epi16(warped, src_pixels);

     // Load 8 elements each of dx and dt, to pair with the 8 elements of dt
     // that we have just computed. Then compute 8 partial sums of dx * dt
     // and dy * dt, implicitly sum to give 4 partial sums of each, and
     // accumulate.
     __m128i dx_row = _mm_loadu_si128((__m128i *)&dx[i * DISFLOW_PATCH_SIZE]);
     __m128i dy_row = _mm_loadu_si128((__m128i *)&dy[i * DISFLOW_PATCH_SIZE]);
     b0_acc = _mm_add_epi32(b0_acc, _mm_madd_epi16(dx_row, dt));
     b1_acc = _mm_add_epi32(b1_acc, _mm_madd_epi16(dy_row, dt));
   }

   // Flatten the two sets of partial sums to find the final value of b
   // We need to set b[0] = sum(b0_acc), b[1] = sum(b1_acc).
   // We need to do 6 additions in total; a `hadd` instruction can take care
   // of four of them, leaving two scalar additions.
   __m128i partial_sum = _mm_hadd_epi32(b0_acc, b1_acc);
   b[0] = _mm_extract_epi32(partial_sum, 0) + _mm_extract_epi32(partial_sum, 1);
   b[1] = _mm_extract_epi32(partial_sum, 2) + _mm_extract_epi32(partial_sum, 3);
 }

 // Compute the x and y gradients of the source patch in a single pass,
 // and store into dx and dy respectively.
 static INLINE void sobel_filter(const uint8_t *src, int src_stride, int16_t *dx,
                                 int16_t *dy) {
   // Loop setup: Load the first two rows (of 10 input rows) and apply
   // the horizontal parts of the two filters
   __m128i row_m1 = _mm_loadu_si128((__m128i *)(src - src_stride - 1));
   __m128i row_m1_a = _mm_cvtepu8_epi16(row_m1);
   __m128i row_m1_b = _mm_cvtepu8_epi16(_mm_srli_si128(row_m1, 1));
   __m128i row_m1_c = _mm_cvtepu8_epi16(_mm_srli_si128(row_m1, 2));

   __m128i row_m1_hsmooth = _mm_add_epi16(_mm_add_epi16(row_m1_a, row_m1_c),
                                          _mm_slli_epi16(row_m1_b, 1));
   __m128i row_m1_hdiff = _mm_sub_epi16(row_m1_a, row_m1_c);

   __m128i row = _mm_loadu_si128((__m128i *)(src - 1));
   __m128i row_a = _mm_cvtepu8_epi16(row);
   __m128i row_b = _mm_cvtepu8_epi16(_mm_srli_si128(row, 1));
   __m128i row_c = _mm_cvtepu8_epi16(_mm_srli_si128(row, 2));

   __m128i row_hsmooth =
       _mm_add_epi16(_mm_add_epi16(row_a, row_c), _mm_slli_epi16(row_b, 1));
   __m128i row_hdiff = _mm_sub_epi16(row_a, row_c);

   // Main loop: For each of the 8 output rows:
   // * Load row i+1 and apply both horizontal filters
   // * Apply vertical filters and store results
   // * Shift rows for next iteration
   for (int i = 0; i < DISFLOW_PATCH_SIZE; i++) {
     // Load row i+1 and apply both horizontal filters
     const __m128i row_p1 =
         _mm_loadu_si128((__m128i *)(src + (i + 1) * src_stride - 1));
     const __m128i row_p1_a = _mm_cvtepu8_epi16(row_p1);
     const __m128i row_p1_b = _mm_cvtepu8_epi16(_mm_srli_si128(row_p1, 1));
     const __m128i row_p1_c = _mm_cvtepu8_epi16(_mm_srli_si128(row_p1, 2));

     const __m128i row_p1_hsmooth = _mm_add_epi16(
         _mm_add_epi16(row_p1_a, row_p1_c), _mm_slli_epi16(row_p1_b, 1));
     const __m128i row_p1_hdiff = _mm_sub_epi16(row_p1_a, row_p1_c);

     // Apply vertical filters and store results
     // dx = vertical smooth(horizontal diff(input))
     // dy = vertical diff(horizontal smooth(input))
     const __m128i dx_row =
         _mm_add_epi16(_mm_add_epi16(row_m1_hdiff, row_p1_hdiff),
                       _mm_slli_epi16(row_hdiff, 1));
     const __m128i dy_row = _mm_sub_epi16(row_m1_hsmooth, row_p1_hsmooth);

     _mm_storeu_si128((__m128i *)(dx + i * DISFLOW_PATCH_SIZE), dx_row);
     _mm_storeu_si128((__m128i *)(dy + i * DISFLOW_PATCH_SIZE), dy_row);

     // Shift rows for next iteration
     // This allows a lot of work to be reused, reducing the number of
     // horizontal filtering operations from 2*3*8 = 48 to 2*10 = 20
     row_m1_hsmooth = row_hsmooth;
     row_m1_hdiff = row_hdiff;
     row_hsmooth = row_p1_hsmooth;
     row_hdiff = row_p1_hdiff;
   }
 }

 static INLINE void compute_flow_matrix(const int16_t *dx, int dx_stride,
                                        const int16_t *dy, int dy_stride,
                                        double *M) {
   __m128i acc[4] = { 0 };

   for (int i = 0; i < DISFLOW_PATCH_SIZE; i++) {
     __m128i dx_row = _mm_loadu_si128((__m128i *)&dx[i * dx_stride]);
     __m128i dy_row = _mm_loadu_si128((__m128i *)&dy[i * dy_stride]);

     acc[0] = _mm_add_epi32(acc[0], _mm_madd_epi16(dx_row, dx_row));
     acc[1] = _mm_add_epi32(acc[1], _mm_madd_epi16(dx_row, dy_row));
     // Don't compute acc[2], as it should be equal to acc[1]
     acc[3] = _mm_add_epi32(acc[3], _mm_madd_epi16(dy_row, dy_row));
   }

   // Condense sums
   __m128i partial_sum_0 = _mm_hadd_epi32(acc[0], acc[1]);
   __m128i partial_sum_1 = _mm_hadd_epi32(acc[1], acc[3]);
   __m128i result = _mm_hadd_epi32(partial_sum_0, partial_sum_1);

   // Apply regularization
   // We follow the standard regularization method of adding `k * I` before
   // inverting. This ensures that the matrix will be invertible.
   //
   // Setting the regularization strength k to 1 seems to work well here, as
   // typical values coming from the other equations are very large (1e5 to
   // 1e6, with an upper limit of around 6e7, at the time of writing).
   // It also preserves the property that all matrix values are whole numbers,
   // which is convenient for integerized SIMD implementation.
   result = _mm_add_epi32(result, _mm_set_epi32(1, 0, 0, 1));

   // Convert results to doubles and store
   _mm_storeu_pd(M, _mm_cvtepi32_pd(result));
   _mm_storeu_pd(M + 2, _mm_cvtepi32_pd(_mm_srli_si128(result, 8)));
 }

 // Try to invert the matrix M
 // Note: Due to the nature of how a least-squares matrix is constructed, all of
 // the eigenvalues will be >= 0, and therefore det M >= 0 as well.
 // The regularization term `+ k * I` further ensures that det M >= k^2.
 // As mentioned in compute_flow_matrix(), here we use k = 1, so det M >= 1.
 // So we don't have to worry about non-invertible matrices here.
 static INLINE void invert_2x2(const double *M, double *M_inv) {
   double det = (M[0] * M[3]) - (M[1] * M[2]);
   assert(det >= 1);
   const double det_inv = 1 / det;

   M_inv[0] = M[3] * det_inv;
   M_inv[1] = -M[1] * det_inv;
   M_inv[2] = -M[2] * det_inv;
   M_inv[3] = M[0] * det_inv;
 }

 void aom_compute_flow_at_point_sse4_1(const uint8_t *src, const uint8_t *ref,
                                       int x, int y, int width, int height,
                                       int stride, double *u, double *v) {
   DECLARE_ALIGNED(16, double, M[4]);
   DECLARE_ALIGNED(16, double, M_inv[4]);
   DECLARE_ALIGNED(16, int16_t, dx[DISFLOW_PATCH_SIZE * DISFLOW_PATCH_SIZE]);
   DECLARE_ALIGNED(16, int16_t, dy[DISFLOW_PATCH_SIZE * DISFLOW_PATCH_SIZE]);
   int b[2];

   // Compute gradients within this patch
   const uint8_t *src_patch = &src[y * stride + x];
   sobel_filter(src_patch, stride, dx, dy);

   compute_flow_matrix(dx, DISFLOW_PATCH_SIZE, dy, DISFLOW_PATCH_SIZE, M);
   invert_2x2(M, M_inv);

   for (int itr = 0; itr < DISFLOW_MAX_ITR; itr++) {
     compute_flow_vector(src, ref, width, height, stride, x, y, *u, *v, dx, dy,
                         b);

     // Solve flow equations to find a better estimate for the flow vector
     // at this point
     const double step_u = M_inv[0] * b[0] + M_inv[1] * b[1];
     const double step_v = M_inv[2] * b[0] + M_inv[3] * b[1];
     *u += fclamp(step_u * DISFLOW_STEP_SIZE, -2, 2);
     *v += fclamp(step_v * DISFLOW_STEP_SIZE, -2, 2);

     if (fabs(step_u) + fabs(step_v) < DISFLOW_STEP_SIZE_THRESOLD) {
       // Stop iteration when we're close to convergence
       break;
     }
   }
 }
	/*
	* Copyright (c) 2024, Alliance for Open Media. All rights reserved
	*
	* This source code is subject to the terms of the BSD 2 Clause License and
	* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
	* was not distributed with this source code in the LICENSE file, you can
	* obtain it at www.aomedia.org/license/software. If the Alliance for Open
	* Media Patent License 1.0 was not distributed with this source code in the
	* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
	*/

	#include <assert.h>
	#include <math.h>
	#include <smmintrin.h>

	#include "aom_dsp/aom_dsp_common.h"
	#include "aom_dsp/flow_estimation/disflow.h"
	#include "aom_dsp/x86/synonyms.h"

	#include "config/aom_dsp_rtcd.h"

	#if DISFLOW_PATCH_SIZE != 8
	#error "Need to change disflow_sse4.c if DISFLOW_PATCH_SIZE != 8"
	#endif

	// Compute horizontal and vertical kernels and return them packed into a
	// register. The coefficient ordering is:
	// h0, h1, v0, v1, h2, h3, v2, v3
	// This is chosen because it takes less work than fully separating the kernels,
	// but it is separated enough that we can pick out each coefficient pair in the
	// main compute_flow_at_point function
	static INLINE __m128i compute_cubic_kernels(double u, double v) {
	const __m128d x = _mm_set_pd(v, u);

	const __m128d x2 = _mm_mul_pd(x, x);
	const __m128d x3 = _mm_mul_pd(x2, x);

	// Macro to multiply a value v by a constant coefficient c
	#define MULC(c, v) _mm_mul_pd(_mm_set1_pd(c), v)

	// Compute floating-point kernel
	// Note: To ensure results are bit-identical to the C code, we need to perform
	// exactly the same sequence of operations here as in the C code.
	__m128d k0 = _mm_sub_pd(_mm_add_pd(MULC(-0.5, x), x2), MULC(0.5, x3));
	__m128d k1 =
	_mm_add_pd(_mm_sub_pd(_mm_set1_pd(1.0), MULC(2.5, x2)), MULC(1.5, x3));
	__m128d k2 =
	_mm_sub_pd(_mm_add_pd(MULC(0.5, x), MULC(2.0, x2)), MULC(1.5, x3));
	__m128d k3 = _mm_add_pd(MULC(-0.5, x2), MULC(0.5, x3));
	#undef MULC

	// Integerize
	__m128d prec = _mm_set1_pd((double)(1 << DISFLOW_INTERP_BITS));

	k0 = _mm_round_pd(_mm_mul_pd(k0, prec),
	_MM_FROUND_TO_NEAREST_INT \| _MM_FROUND_NO_EXC);
	k1 = _mm_round_pd(_mm_mul_pd(k1, prec),
	_MM_FROUND_TO_NEAREST_INT \| _MM_FROUND_NO_EXC);
	k2 = _mm_round_pd(_mm_mul_pd(k2, prec),
	_MM_FROUND_TO_NEAREST_INT \| _MM_FROUND_NO_EXC);
	k3 = _mm_round_pd(_mm_mul_pd(k3, prec),
	_MM_FROUND_TO_NEAREST_INT \| _MM_FROUND_NO_EXC);

	const __m128i c0 = _mm_cvtpd_epi32(k0);
	const __m128i c1 = _mm_cvtpd_epi32(k1);
	const __m128i c2 = _mm_cvtpd_epi32(k2);
	const __m128i c3 = _mm_cvtpd_epi32(k3);

	// Rearrange results and convert down to 16 bits, giving the target output
	// ordering
	const __m128i c01 = _mm_unpacklo_epi32(c0, c1);
	const __m128i c23 = _mm_unpacklo_epi32(c2, c3);
	return _mm_packs_epi32(c01, c23);
	}

	// Compare two regions of width x height pixels, one rooted at position
	// (x, y) in src and the other at (x + u, y + v) in ref.
	// This function returns the sum of squared pixel differences between
	// the two regions.
	//
	// TODO(rachelbarker): Test speed/quality impact of using bilinear interpolation
	// instad of bicubic interpolation
	static INLINE void compute_flow_vector(const uint8_t src, const uint8_t ref,
	int width, int height, int stride, int x,
	int y, double u, double v,
	const int16_t dx, const int16_t dy,
	int *b) {
	// This function is written to do 8x8 convolutions only
	assert(DISFLOW_PATCH_SIZE == 8);

	// Accumulate 4 32-bit partial sums for each element of b
	// These will be flattened at the end.
	__m128i b0_acc = _mm_setzero_si128();
	__m128i b1_acc = _mm_setzero_si128();

	// Split offset into integer and fractional parts, and compute cubic
	// interpolation kernels
	const int u_int = (int)floor(u);
	const int v_int = (int)floor(v);
	const double u_frac = u - floor(u);
	const double v_frac = v - floor(v);

	const __m128i kernels = compute_cubic_kernels(u_frac, v_frac);

	// Storage for intermediate values between the two convolution directions
	DECLARE_ALIGNED(16, int16_t,
	tmp_[DISFLOW_PATCH_SIZE * (DISFLOW_PATCH_SIZE + 3)]);
	int16_t *tmp = tmp_ + DISFLOW_PATCH_SIZE; // Offset by one row

	// Clamp coordinates so that all pixels we fetch will remain within the
	// allocated border region, but allow them to go far enough out that
	// the border pixels' values do not change.
	// Since we are calculating an 8x8 block, the bottom-right pixel
	// in the block has coordinates (x0 + 7, y0 + 7). Then, the cubic
	// interpolation has 4 taps, meaning that the output of pixel
	// (x_w, y_w) depends on the pixels in the range
	// ([x_w - 1, x_w + 2], [y_w - 1, y_w + 2]).
	//
	// Thus the most extreme coordinates which will be fetched are
	// (x0 - 1, y0 - 1) and (x0 + 9, y0 + 9).
	const int x0 = clamp(x + u_int, -9, width);
	const int y0 = clamp(y + v_int, -9, height);

	// Horizontal convolution

	// Prepare the kernel vectors
	// We split the kernel into two vectors with kernel indices:
	// 0, 1, 0, 1, 0, 1, 0, 1, and
	// 2, 3, 2, 3, 2, 3, 2, 3
	__m128i h_kernel_01 = _mm_set1_epi32(_mm_extract_epi32(kernels, 0));
	__m128i h_kernel_23 = _mm_set1_epi32(_mm_extract_epi32(kernels, 2));

	__m128i round_const_h = _mm_set1_epi32(1 << (DISFLOW_INTERP_BITS - 6 - 1));

	for (int i = -1; i < DISFLOW_PATCH_SIZE + 2; ++i) {
	const int y_w = y0 + i;
	const uint8_t ref_row = &ref[y_w stride + (x0 - 1)];
	int16_t tmp_row = &tmp[i DISFLOW_PATCH_SIZE];

	// Load this row of pixels.
	// For an 8x8 patch, we need to load the 8 image pixels + 3 extras,
	// for a total of 11 pixels. Here we load 16 pixels, but only use
	// the first 11.
	__m128i row = _mm_loadu_si128((__m128i *)ref_row);

	// Expand pixels to int16s
	__m128i px_0to7_i16 = _mm_cvtepu8_epi16(row);
	__m128i px_4to10_i16 = _mm_cvtepu8_epi16(_mm_srli_si128(row, 4));

	// Compute first four outputs
	// input pixels 0, 1, 1, 2, 2, 3, 3, 4
	// * kernel 0, 1, 0, 1, 0, 1, 0, 1
	__m128i px0 =
	_mm_unpacklo_epi16(px_0to7_i16, _mm_srli_si128(px_0to7_i16, 2));
	// input pixels 2, 3, 3, 4, 4, 5, 5, 6
	// * kernel 2, 3, 2, 3, 2, 3, 2, 3
	__m128i px1 = _mm_unpacklo_epi16(_mm_srli_si128(px_0to7_i16, 4),
	_mm_srli_si128(px_0to7_i16, 6));
	// Convolve with kernel and sum 2x2 boxes to form first 4 outputs
	__m128i sum0 = _mm_add_epi32(_mm_madd_epi16(px0, h_kernel_01),
	_mm_madd_epi16(px1, h_kernel_23));

	__m128i out0 = _mm_srai_epi32(_mm_add_epi32(sum0, round_const_h),
	DISFLOW_INTERP_BITS - 6);

	// Compute second four outputs
	__m128i px2 =
	_mm_unpacklo_epi16(px_4to10_i16, _mm_srli_si128(px_4to10_i16, 2));
	__m128i px3 = _mm_unpacklo_epi16(_mm_srli_si128(px_4to10_i16, 4),
	_mm_srli_si128(px_4to10_i16, 6));
	__m128i sum1 = _mm_add_epi32(_mm_madd_epi16(px2, h_kernel_01),
	_mm_madd_epi16(px3, h_kernel_23));

	// Round by just enough bits that the result is
	// guaranteed to fit into an i16. Then the next stage can use 16 x 16 -> 32
	// bit multiplies, which should be a fair bit faster than 32 x 32 -> 32
	// as it does now
	// This means shifting down so we have 6 extra bits, for a maximum value
	// of +18360, which can occur if u_frac == 0.5 and the input pixels are
	// {0, 255, 255, 0}.
	__m128i out1 = _mm_srai_epi32(_mm_add_epi32(sum1, round_const_h),
	DISFLOW_INTERP_BITS - 6);

	_mm_storeu_si128((__m128i *)tmp_row, _mm_packs_epi32(out0, out1));
	}

	// Vertical convolution
	const int round_bits = DISFLOW_INTERP_BITS + 6 - DISFLOW_DERIV_SCALE_LOG2;
	__m128i round_const_v = _mm_set1_epi32(1 << (round_bits - 1));

	__m128i v_kernel_01 = _mm_set1_epi32(_mm_extract_epi32(kernels, 1));
	__m128i v_kernel_23 = _mm_set1_epi32(_mm_extract_epi32(kernels, 3));

	for (int i = 0; i < DISFLOW_PATCH_SIZE; ++i) {
	int16_t tmp_row = &tmp[i DISFLOW_PATCH_SIZE];

	// Load 4 rows of 8 x 16-bit values
	__m128i px0 = _mm_loadu_si128((__m128i *)(tmp_row - DISFLOW_PATCH_SIZE));
	__m128i px1 = _mm_loadu_si128((__m128i *)tmp_row);
	__m128i px2 = _mm_loadu_si128((__m128i *)(tmp_row + DISFLOW_PATCH_SIZE));
	__m128i px3 =
	_mm_loadu_si128((__m128i )(tmp_row + 2 DISFLOW_PATCH_SIZE));

	// We want to calculate px0 * v_kernel[0] + px1 * v_kernel[1] + ... ,
	// but each multiply expands its output to 32 bits. So we need to be
	// a little clever about how we do this
	__m128i sum0 = _mm_add_epi32(
	_mm_madd_epi16(_mm_unpacklo_epi16(px0, px1), v_kernel_01),
	_mm_madd_epi16(_mm_unpacklo_epi16(px2, px3), v_kernel_23));
	__m128i sum1 = _mm_add_epi32(
	_mm_madd_epi16(_mm_unpackhi_epi16(px0, px1), v_kernel_01),
	_mm_madd_epi16(_mm_unpackhi_epi16(px2, px3), v_kernel_23));

	__m128i sum0_rounded =
	_mm_srai_epi32(_mm_add_epi32(sum0, round_const_v), round_bits);
	__m128i sum1_rounded =
	_mm_srai_epi32(_mm_add_epi32(sum1, round_const_v), round_bits);

	__m128i warped = _mm_packs_epi32(sum0_rounded, sum1_rounded);
	__m128i src_pixels_u8 =
	_mm_loadl_epi64((__m128i )&src[(y + i) stride + x]);
	__m128i src_pixels = _mm_slli_epi16(_mm_cvtepu8_epi16(src_pixels_u8), 3);

	// Calculate delta from the target patch
	__m128i dt = _mm_sub_epi16(warped, src_pixels);

	// Load 8 elements each of dx and dt, to pair with the 8 elements of dt
	// that we have just computed. Then compute 8 partial sums of dx * dt
	// and dy * dt, implicitly sum to give 4 partial sums of each, and
	// accumulate.
	__m128i dx_row = _mm_loadu_si128((__m128i )&dx[i DISFLOW_PATCH_SIZE]);
	__m128i dy_row = _mm_loadu_si128((__m128i )&dy[i DISFLOW_PATCH_SIZE]);
	b0_acc = _mm_add_epi32(b0_acc, _mm_madd_epi16(dx_row, dt));
	b1_acc = _mm_add_epi32(b1_acc, _mm_madd_epi16(dy_row, dt));
	}

	// Flatten the two sets of partial sums to find the final value of b
	// We need to set b[0] = sum(b0_acc), b[1] = sum(b1_acc).
	// We need to do 6 additions in total; a `hadd` instruction can take care
	// of four of them, leaving two scalar additions.
	__m128i partial_sum = _mm_hadd_epi32(b0_acc, b1_acc);
	b[0] = _mm_extract_epi32(partial_sum, 0) + _mm_extract_epi32(partial_sum, 1);
	b[1] = _mm_extract_epi32(partial_sum, 2) + _mm_extract_epi32(partial_sum, 3);
	}

	// Compute the x and y gradients of the source patch in a single pass,
	// and store into dx and dy respectively.
	static INLINE void sobel_filter(const uint8_t src, int src_stride, int16_t dx,
	int16_t *dy) {
	// Loop setup: Load the first two rows (of 10 input rows) and apply
	// the horizontal parts of the two filters
	__m128i row_m1 = _mm_loadu_si128((__m128i *)(src - src_stride - 1));
	__m128i row_m1_a = _mm_cvtepu8_epi16(row_m1);
	__m128i row_m1_b = _mm_cvtepu8_epi16(_mm_srli_si128(row_m1, 1));
	__m128i row_m1_c = _mm_cvtepu8_epi16(_mm_srli_si128(row_m1, 2));

	__m128i row_m1_hsmooth = _mm_add_epi16(_mm_add_epi16(row_m1_a, row_m1_c),
	_mm_slli_epi16(row_m1_b, 1));
	__m128i row_m1_hdiff = _mm_sub_epi16(row_m1_a, row_m1_c);

	__m128i row = _mm_loadu_si128((__m128i *)(src - 1));
	__m128i row_a = _mm_cvtepu8_epi16(row);
	__m128i row_b = _mm_cvtepu8_epi16(_mm_srli_si128(row, 1));
	__m128i row_c = _mm_cvtepu8_epi16(_mm_srli_si128(row, 2));

	__m128i row_hsmooth =
	_mm_add_epi16(_mm_add_epi16(row_a, row_c), _mm_slli_epi16(row_b, 1));
	__m128i row_hdiff = _mm_sub_epi16(row_a, row_c);

	// Main loop: For each of the 8 output rows:
	// * Load row i+1 and apply both horizontal filters
	// * Apply vertical filters and store results
	// * Shift rows for next iteration
	for (int i = 0; i < DISFLOW_PATCH_SIZE; i++) {
	// Load row i+1 and apply both horizontal filters
	const __m128i row_p1 =
	_mm_loadu_si128((__m128i )(src + (i + 1) src_stride - 1));
	const __m128i row_p1_a = _mm_cvtepu8_epi16(row_p1);
	const __m128i row_p1_b = _mm_cvtepu8_epi16(_mm_srli_si128(row_p1, 1));
	const __m128i row_p1_c = _mm_cvtepu8_epi16(_mm_srli_si128(row_p1, 2));

	const __m128i row_p1_hsmooth = _mm_add_epi16(
	_mm_add_epi16(row_p1_a, row_p1_c), _mm_slli_epi16(row_p1_b, 1));
	const __m128i row_p1_hdiff = _mm_sub_epi16(row_p1_a, row_p1_c);

	// Apply vertical filters and store results
	// dx = vertical smooth(horizontal diff(input))
	// dy = vertical diff(horizontal smooth(input))
	const __m128i dx_row =
	_mm_add_epi16(_mm_add_epi16(row_m1_hdiff, row_p1_hdiff),
	_mm_slli_epi16(row_hdiff, 1));
	const __m128i dy_row = _mm_sub_epi16(row_m1_hsmooth, row_p1_hsmooth);

	_mm_storeu_si128((__m128i )(dx + i DISFLOW_PATCH_SIZE), dx_row);
	_mm_storeu_si128((__m128i )(dy + i DISFLOW_PATCH_SIZE), dy_row);

	// Shift rows for next iteration
	// This allows a lot of work to be reused, reducing the number of
	// horizontal filtering operations from 238 = 48 to 2*10 = 20
	row_m1_hsmooth = row_hsmooth;
	row_m1_hdiff = row_hdiff;
	row_hsmooth = row_p1_hsmooth;
	row_hdiff = row_p1_hdiff;
	}
	}

	static INLINE void compute_flow_matrix(const int16_t *dx, int dx_stride,
	const int16_t *dy, int dy_stride,
	double *M) {
	__m128i acc[4] = { 0 };

	for (int i = 0; i < DISFLOW_PATCH_SIZE; i++) {
	__m128i dx_row = _mm_loadu_si128((__m128i )&dx[i dx_stride]);
	__m128i dy_row = _mm_loadu_si128((__m128i )&dy[i dy_stride]);

	acc[0] = _mm_add_epi32(acc[0], _mm_madd_epi16(dx_row, dx_row));
	acc[1] = _mm_add_epi32(acc[1], _mm_madd_epi16(dx_row, dy_row));
	// Don't compute acc[2], as it should be equal to acc[1]
	acc[3] = _mm_add_epi32(acc[3], _mm_madd_epi16(dy_row, dy_row));
	}

	// Condense sums
	__m128i partial_sum_0 = _mm_hadd_epi32(acc[0], acc[1]);
	__m128i partial_sum_1 = _mm_hadd_epi32(acc[1], acc[3]);
	__m128i result = _mm_hadd_epi32(partial_sum_0, partial_sum_1);

	// Apply regularization
	// We follow the standard regularization method of adding `k * I` before
	// inverting. This ensures that the matrix will be invertible.
	//
	// Setting the regularization strength k to 1 seems to work well here, as
	// typical values coming from the other equations are very large (1e5 to
	// 1e6, with an upper limit of around 6e7, at the time of writing).
	// It also preserves the property that all matrix values are whole numbers,
	// which is convenient for integerized SIMD implementation.
	result = _mm_add_epi32(result, _mm_set_epi32(1, 0, 0, 1));

	// Convert results to doubles and store
	_mm_storeu_pd(M, _mm_cvtepi32_pd(result));
	_mm_storeu_pd(M + 2, _mm_cvtepi32_pd(_mm_srli_si128(result, 8)));
	}

	// Try to invert the matrix M
	// Note: Due to the nature of how a least-squares matrix is constructed, all of
	// the eigenvalues will be >= 0, and therefore det M >= 0 as well.
	// The regularization term `+ k * I` further ensures that det M >= k^2.
	// As mentioned in compute_flow_matrix(), here we use k = 1, so det M >= 1.
	// So we don't have to worry about non-invertible matrices here.
	static INLINE void invert_2x2(const double M, double M_inv) {
	double det = (M[0] * M[3]) - (M[1] * M[2]);
	assert(det >= 1);
	const double det_inv = 1 / det;

	M_inv[0] = M[3] * det_inv;
	M_inv[1] = -M[1] * det_inv;
	M_inv[2] = -M[2] * det_inv;
	M_inv[3] = M[0] * det_inv;
	}

	void aom_compute_flow_at_point_sse4_1(const uint8_t src, const uint8_t ref,
	int x, int y, int width, int height,
	int stride, double u, double v) {
	DECLARE_ALIGNED(16, double, M[4]);
	DECLARE_ALIGNED(16, double, M_inv[4]);
	DECLARE_ALIGNED(16, int16_t, dx[DISFLOW_PATCH_SIZE * DISFLOW_PATCH_SIZE]);
	DECLARE_ALIGNED(16, int16_t, dy[DISFLOW_PATCH_SIZE * DISFLOW_PATCH_SIZE]);
	int b[2];

	// Compute gradients within this patch
	const uint8_t src_patch = &src[y stride + x];
	sobel_filter(src_patch, stride, dx, dy);

	compute_flow_matrix(dx, DISFLOW_PATCH_SIZE, dy, DISFLOW_PATCH_SIZE, M);
	invert_2x2(M, M_inv);

	for (int itr = 0; itr < DISFLOW_MAX_ITR; itr++) {
	compute_flow_vector(src, ref, width, height, stride, x, y, u, v, dx, dy,
	b);

	// Solve flow equations to find a better estimate for the flow vector
	// at this point
	const double step_u = M_inv[0] * b[0] + M_inv[1] * b[1];
	const double step_v = M_inv[2] * b[0] + M_inv[3] * b[1];
	u += fclamp(step_u DISFLOW_STEP_SIZE, -2, 2);
	v += fclamp(step_v DISFLOW_STEP_SIZE, -2, 2);

	if (fabs(step_u) + fabs(step_v) < DISFLOW_STEP_SIZE_THRESOLD) {
	// Stop iteration when we're close to convergence
	break;
	}
	}
	}